The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external sources and has significant write capabilities.
Ingestion points: The skill uses Firecrawl to scrape target and competitor web pages into markdown (references/firecrawl.md) and uses DataForSEO to gather SERP and AI visibility data (references/dataforseo.md).
Boundary markers: The skill includes a verification step in references/apply-loop.md to ensure "no internal-rule/provenance phrases are visible" in the output, which serves as a basic check against accidental obedience to injected instructions.
Capability inventory: The skill allows the agent to perform content CRUD operations and "Astro code patches" to modify the site's rendering and source code (SKILL.md, references/apply-loop.md).
Sanitization: The workflow relies on a human-in-the-loop (HITL) model and verification steps to inspect rendered output and consistency, though it lacks automated sanitization of the scraped markdown content before it influences code generation.
[COMMAND_EXECUTION]: The skill instructs the agent to use the dineway CLI tool and associated subcommands (e.g., dineway seo firecrawl, dineway seo dataforseo) to interact with hosted providers. It also authorizes the agent to propose and apply "Astro code patches" to the project's codebase when standard CMS fields are insufficient for SEO requirements.
[EXTERNAL_DOWNLOADS]: The workflow involves fetching data from remote URLs via third-party SEO providers. While these are routed through a vendor-controlled API base (DINEWAY_API_BASE_URL), the content returned (web scrapes and SERP data) is external and untrusted.

dineway-seo